Kinocut MCP
Local video editing for agents: typed tools over FFmpeg, quality checks and a provenance receipt before you publish
- 251 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- Runs FFmpeg on your files and writes results to disk, and overwriting existing outputs needs an explicit `overwrite` parameter
- Optional modules launch Hyperframes, Whisper and local models, and the agent's request may reach its model provider
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/kinocutDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Runs FFmpeg on your files and writes results to disk, and overwriting existing outputs needs an explicit `overwrite` parameter
- Optional modules launch Hyperframes, Whisper and local models, and the agent's request may reach its model provider
This is third-party code. Review the repository files before installing.
What it does
Kinocut gives an agent a set of typed video operations instead of guessing FFmpeg flags. The agent trims, merges, resizes, adds subtitles and text, normalizes audio, and makes thumbnails and storyboards. Each operation is validated before it runs, and the result comes back as a structured answer with an output path and a Video Receipt that records inputs, hashes and stages. Everything runs on your computer, with no cloud editor and no account. The base package offers three ways to work: an MCP server (`kino --mcp`), the `kino` CLI with 177 commands and a Python `Client`. By default the server registers 20 core tools, while `KINOCUT_TOOLS=all` opens all 205, and hidden operations are found with `search_tools`. For safe publishing there are `video_quality_check` and `video_release_checkpoint`. A person should still watch the finished video before release. Ready-made scenarios sit on top: `video_shorts` builds a vertical captioned clip in one call, repurposing turns one recording into variants for Shorts, Reels and TikTok, and the workflow engine runs a JSON description of a chain and writes a receipt. Separate modes assemble a two-camera edit from stitched 360 video and rescue damaged clips. The repository carries two skills, `kinocut` and `kinocut-repurpose`. The project used to be called mcp-video.
Who it is for. Podcast, interview and short-video makers who want to hand routine editing to an agent and keep control of quality before release.
Good fit when
- An interview recording should become a 45-second vertical clip with burned-in captions and normalized audio
- One podcast is needed in three formats for Shorts, Reels and TikTok with a manifest and thumbnails
- The agent assembled an edit and before release you want a quality report and a receipt with hashes
- A chain of trimming, resizing and titles should be described in JSON and repeated in CI
- A dual 360 camera shoot turns into an edit with switching, picture-in-picture or a split screen
Not a fit when
- You need manual timeline editing with frame tuning: Kinocut is driven by commands and does not replace a graphical editor
- Rendering must run in the cloud on someone else's servers: all operations run locally
- The source is in `.insv` or `.360` format: Kinocut rejects it, so export a stitched MP4 first
Example request
Trim interview.mp4 to the strongest 45 seconds, add burned-in captions, make it 9:16 vertical and run a quality check before exportMCP
- Transport
- stdio
- Authentication
- not required
Environment variables
KINOCUT_TOOLS- Tool set: `all` opens all 205, `profile:editor` takes the editor profile
MCP_VIDEO_HYPERFRAMES_COMMAND- Hyperframes launch command if it is not on `PATH`
Limitations
- Python 3.11 or newer and FFmpeg on `PATH` are required. Hyperframes tools additionally need Node.js 22 and the Hyperframes command line.
- Optional extras are heavy: Whisper transcription is about 1 GB, stem separation about 2 GB, upscaling about 2 GB and only on Python up to 3.12.
- By default the agent sees 20 of 205 tools. The rest are enabled with the `KINOCUT_TOOLS` variable or found through `search_tools`.
- Some capabilities are marked in the README as gated or partial, for example the unsigned MCPB package for Claude Desktop.
How to disable. Remove the server with `claude mcp remove kinocut` or delete the `kinocut` entry from the client's MCP configuration. Uninstall the package with `pip uninstall kinocut`. Generated videos and receipts stay in working directories and are deleted by hand.
FAQ
How do I connect Kinocut to Claude Code?
Install FFmpeg and run `claude mcp add kinocut -- uvx --from kinocut kino --mcp`. You can check the environment with `uvx --from kinocut kino doctor`.
How do I add Kinocut to Cursor or Claude Desktop?
Add a `kinocut` server to the MCP configuration with the command `uvx` and the arguments `--from kinocut kino --mcp`. Any MCP client that can launch a local stdio server will do.
Why does the agent see only 20 tools?
That is the curated default set. Set `KINOCUT_TOOLS=all` to see all 205, or `KINOCUT_TOOLS=profile:editor` to take the editor profile.
Does it need an API key or an account?
Not for the core: Kinocut runs locally and does not upload video to a Kinocut cloud. You pay only for optional providers and for the model inside the agent, and prompts may reach that model's provider.
What should mcp-video users do?
Update the compatibility shim with `pip install --upgrade 'mcp-video==1.6.20'` and then `pip install --force-reinstall --no-deps 'kinocut==1.16.5'`. Old imports, the `mcp-video` command and the `MCP_VIDEO_*` variables keep working.
Discussion
No questions or reviews yet.
Related
MCP serversMCP server, CLI and desktop app that cut a long video or stream replay into vertical clips with captions
An open-source video editor with a Codex plugin: the agent opens a project in the side panel, imports media, edits in batches
A video-director skill: from one sentence it builds a project with footage, storyboard, captions, motion, an ffmpeg render and quality checks
ffmpeg-skill — a skill and MCP server: an agent cuts and joins video, fixes audio, adds captions and exports for TikTok and Reels with local FFmpeg
A Claude Code skill that turns a talking-head video into a short motion-graphics edit and checks its own cut
chengfeng-videocut
chengfeng-videocut-skills
An agent plugin for editing talking-head video: removing verbal slips, subtitles, on-screen graphics and rendering the final cut
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Kinocut (this card) | no | not required | 251 | |
| ffmpeg-skill | no | 4,619 | a skill and MCP server over local FFmpeg for TikTok and Reels edits; Kinocut adds result receipts, quality checks and JSON workflows | |
| video-use | no | 30,591 | a skill that cuts pauses, grades color and burns subtitles; Kinocut offers typed tools, an MCP server and a CLI with 177 commands | |
| HyperFrames | no | 79,556 | draws videos from HTML compositions; Kinocut edits ready video files and plugs Hyperframes in as an optional module |
README in short
The document opens with a brief definition and a table of project facts. It continues with call examples, a release history with tool counts, workflow and 360-video scenarios, installation with an extras table, a quick start, a tool list and answers to common questions.